Chip-seq on pooled replicates

Question

[GR]

Dear All,

Can someone please help me to understand why for ChIP-seq we should not merge the biological replicates before peak calling. I read in several papers, peaks calling is done on individual replicates. I am struggling to understand this. Ideally if I merge replicates I should have more power, means more coverage in peak regions.

My results also indicates that if I call peaks on individual replicates, I get around 10,000 peaks in each rep. I have 3 replicates for chip and 3 for control. If I merge these together and run macs2, I get very few peaks < 100. Can anyone please explain this.

Thanks. RT

[cmdcolin]

I think sometimes replicates do just get pooled so it's not out of the question, and many fewer programs exist for properly handling chip-seq replicates separately. the IDR pipeline from encode tried to tackle the problem ("Irreproducibility Discovery Rate") but it is probably just good to manually inspect the data like @EagleEye recommends to see if there are problems with the enrichment or something.

Answer 1

[EagleEye]

Before merging the replicates please check how the replicate samples/experiments correlated (example, PCA, deeptools etc.,).

Compare replicates to check their correlation

[datascientist28]

Yeah it seems that the noise in a sample could be adding to this.